GPU physics works, kinda

I remember a while back there was the suggestion that in a two or three slot PCI-E system you could use your old graphics card for physics while your new graphics card(s) did the graphics. I can't remember if this was an ATI or Nvidia idea (or both).

I'm not sure I like the idea of off-loading physics to the same GPU that is also doing the graphics. However, being able to stuff an old or low end card into the system to offload physics would be ideal. Hopefully this idea is still on the cards.

What this and the original article forget to mention is that even in UE3 the primary phsyics is still always handled by the CPU, and it uses Epic's own physics engine at the core, just like GRAW the PhysX portion is added to another physics engine (Havoc is used for GRAW) and then the PhysX engine is just fluff ontop.

Until it's a ground-up addition, it's like adding 16XAA instead of 4XAA and talking about that performance as if it's something everyone should care about. 

What do you wanna bet that the 'too stressed to do PhysX calculations' is reached at a much lower level on a GF8600, which is the majority of users not the GF9800GTX, and the majority of those mid-range users would be the ones looking for free boost instead of buying something more for the boost. The 9800GTX user likely would have no trouble forking out some more money for a card to give a hybrid SLi or PhysX boost.

But most importantly as it related to the Vantage fiasco, it shows that compared to a dedicated PPU it can't compete in real games since it need to split the workload whereas in Vantage it performs better than a PPU because it goes from 100% GPU to 100% PPU mimicry without a need for the balance that games require. Once again proving that 3Dmark's value is lessened by a comparison to real-world results.

It would be interesting to see how the performance looks on a 2 GPU system, say, on an Intel P35 Board with 2 PCI Express Slots and one Geforce 8800GT and a Radeon 4700.

Teh Radeon gets teh graphics done and teh Geforce crunches teh PhysX. AFAIK, only G92 are supported as of yet. it would be *really* neat, if one could use an 8600GT or something similar to get PhysX done while having teh graphics processed by a completely different card, such as a Radeon or a 7900GTX.

Any experiences, anyone?

In theory, you'd be able to use SLI (or potentially crossfire) to power a physics engine (e.g. PhysX) using one of the graphics cards. This would be particularly interesting with these hybrid configurations, using integrated graphics for physics and your main card(s) for graphics.

THe problem is the physic api is designed to work best on hardware, not necessarily great hardware, just specific hardware setup a specific way. The software, being owned by Nvidia now, doesn't have to be fully optimised to run on the cpu, it can simply run ok, but be fully optimised to run well on anything that well, makes it money.

Then you brought up the question does Physx bring anything to gaming? no not at all, realistic physics calculations has entirely nothing to do with realistic LOOKING physics. I human eye can't actually tell the difference between a dead body falling with a 10m/s acceleration and a 9.8m/s acceleration, but one rounds up into easier numbers that can be calculated quicker. This is the problem, the idea is that more realistic physics CALCULATIONS directly leads to improved physics in games, this would only be true if minute changes would be perceptable to the human eye, they aren't at all. The use of physic, ,accurate or estimated in games give the same end product, it doesn't matter which you use but adding more detailed physics effects, even cheap to calculate ones improves the feel of the game. Physx/nvidia will have you believe this is only possible with super accurate calculations but all evidence points to it simply being a design time problem. Games have always increased the level of physics use in games, from falling more accurately, ragdoll movement of enemies, to being able to see enviroments explode. This has all been increased without accurate physics, it just takes longer to design a wall that explodes, clips correctly and doesn't bring about bugs or screw up the multiple scripted attacks in games when enemies get stuck in a wall thats half destroyed. Physx is infact making it HARDER to introduce these things into games as by making the calculations accurate you are actually making it a longer and harder design process.

I thought it would be nice if the physics accerlation in UT3 can be pulled by nvidia DX10 cards (if not all, at least the mid/high-end ones), so players of UT3 have a bigger chance of play the game online at HD-physx enabled servers. users of the dedicated ageia PPU PCI card are not comparable in number to users of high end nvidia cards. thanks

They do not want to give any cycle away from the GPU, they want to use all the cycles for rasterization related work, make it more realistic by doing more multi layer texturing, more shaders. Somebody is not listening to its customers...

who?

I use a 24" 1920x monitor and a 9800GTX :)

And honestly that is a bare minimum for that kind of resolution. So even I have a need for little more performance boost, not just the "mainstream" people.

No really WTF is a "HD-physx enabled server".

Some commentators here specialise in speaking made up crap pulled directly out of their arses, you know.. the type where you brush the sh*t off of it, and serve it up as gospel truth....

A bit like an Nvidia press release I guess....

and about as valid as an NVidia Futuremark score. 

Aah well...

Oh and by the way everyone else is at DX10.1, DX10 is soooo last quarter..


Original Aegia released the PPU to take up the calculations of PhysX away from the CPU to help improve FPS in games that would use their API. Moving cloths/more debri etc. They sold roughly 100,000 units and quite a few games have come out since using the PPU. These games still have the basic physics in there the concept of Aegia was to allow for more with less of a hit on the FPS. In comes Nvidia who purchases Aegia and convert it into a software that would enable the same features as the PPU and would gain access to millions of clients which would be more enticing for game developers & others to implement PhysX into there products. Nvidia has done it's part and has shown that it can have PhysX on the GPU. Now for sure running it off one GPU to split graphics/PhysX you are possibly going to get a hit in performance at higher resolutions but it probably wouldn't be worse then the basic physics being run off the CPU. The real benefit will probably come from dual video setups were one card will do GPU and the other will do PhysX. I think Nvidia needs to clarify exactly what they intended with there own words as there seems to be alot of confusion for nothing.

People that run extreme resolutions mostly have SLI-setups since no single nv-card can do 2K-resolutions with high settings without grinding to a halt. With two or three SLI'd GPU's there's no problem spending 20% of the resources on one of those cards on physx, the extra physx-power will leverage the drop in GPU-power and then some. With 8800GT's being dirt cheap it's almost conceivable to buy an extra 8800GT just to use as a physx accelerator. That would even work on non-SLI Intel-based motherboard (minus the extra latency of sending data between the two cards on the PCIe-bus since the calculations aren't done on the same card). However, nVidia being greedy and all that I doubt they'll allow something like that to work even though it's not a matter of SLI. We'll see how the whole physx-project finalizes.

Ultie likes to Gar Goon numbers, yet its way to make thinkers think. Here PhysX numbers are beginning to look Reliable.

Obvious question is Why isn't Nvidia puting out 280 GTX2 with dual PhysX PPU in it & Question, How about Slew of PCI PhysicsX cards, in spite of fact off market, maybe that was hasty decision. 
Four PhysX pci cards 2 PhysX on dual Gpu Nvidia card Plus two or three of Em, Just for Blast. SEE IF YOUR NUMBER MACHINE GOES THAT HIGH, Msr. OberPricer.

Signed:Cheap & Powerful.

i'm not sure if you are aware..to play an agiea PhysX game online, with PhysX enabled. all ends need to have the PPU hardware accelerator running, as well the game server must be set to HD-Physics enabled (it's mentioned like that in the UT3 menu, i did not come up with that!).. so in short im saying, with the currently large users of nvidia DX10 cards, it's would be easier for HD-Physics UT3 servers to exist than the case is with the very few Ageia PPU users.

Note the game server itself need to have HD-Physics enabled. it could be the ageia PPU PCI-card which is known for low sales, or (if it actually happens) the dx10 cards which are used by million of gamers world wide.

I hope you got the point.
It could be the same with any physx title. old one, currently in development or ones coming in the future. 

That also could boost sales of UT3 title and maybe other games. wouldn't find that surprise. gamers love trying new things, they rarely get to do so.